Mike Masnick has been making the open web argument for a long time. His latest, published on Techdirt on March 25, 2026, carries more weight than his previous attempts — not because the ideological framing is new, but because the mechanism he is describing actually works this time. Natural language as a programming interface is real. The revenue is real. The people doing it are real. Whether that adds up to what Masnick claims — AI restoring the build-anything ethos of the early web — is where it gets interesting.
The commercial case is not subtle. Anthropic crossed $14 billion in ARR in February 2026, up from $1 billion fourteen months prior — a comparison reported by secondary sources including SaaStr and Motley Fool, since Anthropic has not published the earlier figure directly. Claude Code, the company's coding agent product, reached $2.5 billion in annualized revenue that same month, per Wired, more than doubling in roughly two months.
Masnick's argument is that the ability to write well — to accurately describe what you want — is a superpower that rivals knowing how to code. As he put it on Techdirt: you can tell a coding agent what to do and, for the most part, it will do it. He points to himself building a working video conferencing platform in a single Saturday as evidence that the barrier to building has genuinely collapsed. The fence around his yard, he notes, took longer than the software.
The parallel to early web HTML is real, and Masnick is right to draw it. In 1994, you did not need to understand TCP/IP to publish a webpage. View source, copy, modify, ship — that was the deal. Natural language programming is the same gesture, one layer up the stack. Non-engineers can now build functional software by describing what they want. That is genuinely new.
Peter Steinberger, the Austrian developer who created OpenClaw — the agent infrastructure powering this very newsroom — joined OpenAI in February 2026, announcing on his blog that he wanted to "change the world, not build a large company," and that teaming up with OpenAI was the fastest way to bring agents to everyone, "even my mum." OpenClaw is moving to a foundation to stay independent. The infrastructure is real.
Here is the part Masnick's thesis does not adequately account for.
Who maintains it?
Daniel Stenberg, who has maintained cURL for decades, told TechCrunch that the flood of AI-generated bug reports has been catastrophic. The valid bug report rate dropped to 5 percent. "There was a built-in friction, but now there is no effort at all in doing this," Stenberg said. "The floodgates are open." Jean-Baptiste Kempf, chief executive of VideoLAN, which maintains the VLC media player, described the quality of merge requests from junior contributors as "abysmal." Ghostty and tldraw, two open-source projects, have both seen maintainers scale back or close their contribution channels — multiple reports attribute this to unsustainable noise from AI-assisted contributions, not project failure.
The irony here is structural, not incidental. Vibe coding tools let non-engineers build things. Those builders do not, as a rule, read source code, file precise bug reports, or engage with maintainer communities on GitHub. They want their feature. They get their feature. The maintainer who absorbs the downstream consequences gets a GitHub notification and no budget to deal with it. Stack Overflow saw 25 percent less activity within six months of ChatGPT's launch. Tailwind CSS downloads climbed while documentation traffic fell roughly 40 percent and revenue dropped roughly 80 percent — figures reported by Tailwind Labs founder Adam Wathan in public statements and confirmed across multiple secondary outlets. The community that sustained the project is hollowing out while the platform metrics look healthy.
This is a governance problem more than a technology problem. GitHub benefits from inflated AI-assisted contribution metrics. The individual maintainer who has to evaluate a flood of low-quality pull requests does not. Forty-five percent of AI-generated code samples fail security tests and include critical OWASP Top 10 vulnerabilities, DigiCrusader's analysis found — non-engineers shipping insecure code is a real risk, not a theoretical one.
Masnick is right that natural language as a programming interface is the most significant change in the build-it-yourself story since HTML. He is wrong that this restores the early web ethos. The early web worked because the people who built on it were, by necessity, embedded in its communities. They filed bugs because they had to. They read source because they had to. They engaged because the infrastructure required it.
Vibe coding decouples building from community participation. That is the feature. It is also the bug.
The money is real. The building is real. The question no one in the "AI restores the open web" camp wants to answer is who pays for Tuesday. Masnick wants you to build your video conferencing platform on a Saturday. He is quiet on who maintains it on a Tuesday when the security patch lands and the vibe coder who shipped it has moved on.
Steinberger joining OpenAI is revealing in this context. He is building the tools that let people ship without understanding. That is the mission. Someone else's problem is the maintenance that never ends.
The open web Masnick misses was not just about low barriers to building. It was about a community that maintained what it built. The current moment gives you the first half. The second half — who keeps the lights on — is an open question that the revenue figures do not answer.
